Networks in Construction Grammar

In the last 15 years or so, Construction Grammar has established itself firmly in linguistics. It meshes particularly well with usage-based approaches (see Bybee and Beckner 2010: 842-845), especially in the fields of variational linguistics, diachronic linguistics and language acquisition. Inevitably, the scope of what is understood by Construction Grammar has widened, so that it is nowadays more accurate to speak of Construction Grammars - plural - than of Construction Grammar (see Croft and Cruse 2004).

Some versions of Construction Grammar have added epithets to the name, as for instance Radical Construction Grammar (Croft 2001), Fluid Construction Grammar (Steels 2011b), Embodied Construction Grammar (Bergen and Chang 2013) or Sign-Based Construction Grammar (Michaelis 2009). Not all of these Construction Grammar variants share the same views on the ins and outs of the model, and they disagree on the technical representation. Still, there is a core creed to which most of them subscribe. Language is basically a fund of “constructions”, pairings of form and meaning, of varying degrees of complexity, and these constructions form taxonomic networks. The vertical dimension of these networks revolves around the idea of schematicity: the higher positions in the network are occupied by schematic constructions, which hierarchically subsume the lower positions with (partially) specific instantiations of the abstract constructions. Whereas more traditional linguistic approaches have a division between syntax and the lexicon, constructional approaches to language generally reject such a division (see e.g. Croft and Cruse 2004: 255-256). In the words of Goldberg (2006:18): “It’s constructions all the way down”.[1] Indeed, if schematic - that is lexically underspecified - constructions like the ditransitive for instance have their own idiosyncratic meaning that cannot be reduced to its parts (see Goldberg 1995), they do not differ really from individual words, as they realise an irreducible form-function correspondence. The same applies to the level below the word: in Construction Grammar bound morphemes convey meaning in a way not unlike free morphemes (see Booij 2010 in defence of Constructional Morphology; also see Booijand Huning, this volume).[2] The various types of constructions can be classified according to the dimension atomic- complex and the dimension schematic-specific, which are orthogonal to each other. As becomes clear in Table 1, the latter dimension is gradual in nature: partially schematic atomic constructions are bound morphemes and partially schematic complex constructions are multi-word constructions with a mixture of lexically fixed parts and lexically underspecified slots.

Table 1: Different types of constructions


Partially schematic







[Transitive NP V NP]

[V [Poss way] PP]

let alone

Constructions of different sizes, from fully lexically instantiated multi-word expressions (e.g. kick the bucket) over partially lexically instantiated constructions like the way-construction (see Goldberg 1996; Israel 1996), over bound morphemes (e.g. the third person verbal ending -s) to fully lexically underspecified constructions (e.g. the ditransitive construction), form a giant network, and each node in this network has its own features that specify the meaning import of that particular construction. The whole network of constructions in a particular language is called the Constructicon (Evans 2007: 42), and is in essence an extended version of what other theories would regard as the Lexicon, enriched with non-lexical constructions.[3]

Along the vertical schematic-specific dimension in the Constructicon the lower concrete constructions “are sanctioned by” and “inherit features from” the higher schematic construction, and features from lower nodes in the network “percolate” upward to the higher nodes. An example of a - partial and simplified - hierarchical network is shown in Figure 1. The top node is the maximally underspecified “transitive” construction. At the next level there are nodes for the ditransitive, the transitive resultative and the reflexive construction. These three constructions are all sanctioned by the transitive construction, but in different ways: the ditransitive is an “extension” of the transitive construction with an extra participant, the transitive resultative is a blend of the transitive and a resultative predicate, and the reflexive is a lexically more specific instantiation of the transitive, in which the direct object Noun Phrase is filled with the reflexive pronoun. Fully instantiated versions of the ditransitive and the transitive resultative constructions are examples like Hij heeft haar een brief gestuurd (‘He sent her a letter’) and Hij kuste haar bewusteloos (‘He kissed her unconscious’), respectively. At the same time, there are other nodes in the hierarchy that represent constructions that draw on different parent constructions. One of these is a construction that may be called “fake object resultative construction”, an instantiation of which is Hij loopt zijn schoenen stuk (lit. ‘He walks his shoes broken’). In actual fact, it is not really a separate construction next to the transitive resultative construction, but an extension of it by inserting a normally intransitive verb (lopen ‘walk’) in a resultative construction frame, by a process called “coercion”. This process shows the power of Construction Grammar, which has fewer difficulties with such constructions than traditional grammar. The intermarriage between the reflexive and the fake object resultative yields a construction which for convenience sake has been termed “fake reflexive resultative” in the network in Figure 1. Constructs instantiating it are of the type Hij schreeuwt zich schor (lit. ‘He shouts himself hoarse’). The offshoot of the ditransitive construction and the reflexive is the ditransitive reflexive (Ronny gunt zich een verzetje, lit. ‘Ronny grants himself a distraction’), and the combination of this ditransitive reflexive with the caused motion construction gives us the Dutch weg-construction, which is similar to the English way-construction, but has a somewhat different form (see Verhagen 2002, 2003).

A hierarchical network of constructions in Dutch

Figure 1: A hierarchical network of constructions in Dutch

The kind of hierarchical network of constructions with inheritance relations illustrated in Figure 1 is well-known, and features in various scholarly articles on Construction Grammar. There is, however, another respect in which constructions form a network: rather than just forming a hierarchical structure, constructions can also be related to each other on what could be called the horizontal axis.[4] What I have in mind here is a network where the form-function relation of a particular construction may be partly motivated in relation to its neighbours. This view on networks is familiar from phonology and morphology, but is less often applied to syntax. Before discussing syntactic examples in section 2, I will first elucidate the horizontal relations in a network in phonology and morphology.

In phonology, segmental elements can be related to abstract nodes higher up in the hierarchy thus forming a network not unlike the type illustrated in Figure 1. In Dutch, /е/ or /э/ are instantiations at the lowest level of the higher abstract node “short front vowel” and “short back vowel”, respectively. The latter nodes straightforwardly instantiate nodes of an even higher level. This is all illustrated in Figure 2.

Partial hierarchical network in Dutch phonology

Figure 2: Partial hierarchical network in Dutch phonology

The hierarchical network can be enriched by adding lines for horizontal relations. There is a relationship between the two vowels at the lowest level:

they share the features [+vowel] and [+short] and are differentiated by the feature [± front] (and also [± rounded], but this is ignored here in order not to complicate the issue unnecessarily). It is not just the vowels themselves that entertain this horizontal relation, but the abstract nodes as well. The boxes for front vowels and back vowels are not just arbitrary notions, but are in a contrastive set ([+ front] and [- front]), and the different values impact on the meaning. By contrast, the nodes [reflexive], [ditransitive] and [resultative] in Figure 1 do not stand in such a contrastive relation: there is no sensible interpretation of [+reflexive] as [- ditransitive]. In Figure 3, double-headed arrows are added to bring out the horizontal relationships in the phonological network.

Partial hierarchical network in Dutch phonology with horizontal relations added

Figure 3: Partial hierarchical network in Dutch phonology with horizontal relations added

In morphology, horizontal relations in networks are well-known, and are mostly referred to as “paradigms”. The Dutch verb forms spreek (speak.1SG), spreekt (speak.2/3SG), spreken (speak.PL) are all connected in a large network, but their mutual relation is, again, not one of instantiation. Rather they are at the same hierarchical level and are horizontally differentiated by the inflectional endings.[5] The horizontal relations in the constructional network in Figure 4 are again visualised by double-headed arrows. The existence of similar “paradigms” outside conjugation or declination has received less attention. Nonetheless, horizontal relations in constructional networks can be discerned at the level of syntax as well, as will be shown in the next section.

A hierarchical network with horizontal relations added

Figure 4: A hierarchical network with horizontal relations added

  • [1] There is disagreement about whether fully instantiated constructions (also called “constructs”)also count as constructions (see Taylor 2004 for a critique; see also Norde, De Clerck, andColleman, this volume).
  • [2] This is not a new idea, but goes back to at least Bloomfield, and it is one of the foundationalprinciples of the Columbia School of Linguistics, as set out by Diver (Huffman 2001). Van derHorst (1995: 239) interestingly points out that there seems to be an iconic relation betweenform and meaning such that concrete, lexical meaning is encoded by concrete lexemes,whereas bound morphemes and even less “material”, non-segmental signifiants (e.g. word orderregularities such as V2 (verb-second) or dependent-before-head) have a more organising signifie.In other words: it would be strange to have a language that expresses a meaning like ‘cat’ or‘table’ with a bound morpheme, let alone with a non-segmental form, and it would be equallystrange to mark illocution with a lexical verb or an auxiliary (see Hengeveld 2004:1198-1199).See Van de Velde (2009: 144) for the incorporation of this idea in grammaticalisation theory.
  • [3] The term constructicon is said to be originally coined by Jurafsky (1992).
  • [4] Taylor (2004) also embraces the idea of constructions forming networks: “(...) each unitstands at the hub of a network of relations to other units” (2004: 49) and speaks about horizontalrelations in constructions, but he refers to something else, viz. the relationship between a structure and a larger structure of which it is a part, e.g. the relation between a Noun Phrase andthe clause in which the Noun Phrase functions as, say, a direct object. By contrast, horizontalnetworks in this paper refer to structures of differential relationships between NPs functioningas subjects, direct objects, indirect objects etc. (see section 2.2).
  • [5] The situation is complicated by the existence of homonymous forms: spreek is also the imperative form of the verb spreken and spreken can be an infinitive as well as a plural form. Moreover,with post-verbal subjects, the form for 2SG is not spreekt but spreek. These complications willbe ignored in the present paper, where the morphological example merely has an expositoryfunction.
< Prev   CONTENTS   Source   Next >